抄録
This paper considers efficiency of a hit-or-miss Monte Carlo (MC) method in which each relevant event is simulated on computer. For instance, a failure probability of a system is evaluated as a relative frequency of the failure state occurrence. The method is not efficient, but can be easily applied to various problems without restriction. The method is also robust. Therefore, consideration on its computational efficiency is significant. This paper proposes two methods to improve the efficiency of the above method for structural reliability evaluation. One method is called an Uncommon-Events-Sampling (UES) method in this paper. This method calculates a relative frequency of the failure state occurrence by using nothings but samples corresponding to uncommon events that are significant for reliability evaluation. Next, by considering a ratio of this relative frequency and the required failure probability, this method evaluates the failure probability. Another method is called an Assigning-Pseudorandom-Numbers (APN) method. Although a general MC method uses pseudorandom numbers in their generating order, the proposed method assigns the pseudorandom numbers to basic variables according to their relationship with which the failure state easily occurs. In a numerical example, reliability analysis of a steel truss bridge designed based on Japanese current code is carried out by using the proposed methods. Effectiveness of the proposed methods is illustrated through the numerical example.
